Is there any free library to let a user easily build a C mathematical expression, which can be used like any other function? I mean c expression/function which could be as quick as 'inline' mathematical expression and could be used many times in program.
I think I can be done in C somehow, but does anybody can tell if it could be real if it have to be a CUDA dveice function?
There are a few options. I assume you want something that the user can "call" several times, like this:
void *s = make_func("2 * x + 7");
...
printf("%lf\n", call_func(s, 3.0)); // prints 13
...
printf("%lf\n", call_func(s, 5.0)); // prints 17
...
free_func(s);
One option is to implement this as a recursive structure holding function pointers and constants. Something like:
enum item_type { VAR, CONST, FUNC };
struct var {
enum item_type;
int id;
};
struct constant {
enum item_type;
double value;
};
struct func {
enum item_type;
double (*func)(double, double);
enum item_type *a, *b;
};
Then make_func would parse the above string into something like:
(struct func *){ FUNC, &plus,
(struct func *){ FUNC, ×,
(struct constant *){ CONST, 2 },
(struct var *){ VAR, 'x' } }
(struct constant *){ CONST, 7 } }
If you can understand that - the enum type_item in the struct func is used to point to the next node in the tree (or rather, the first element of that node, which is the enum), and the enum is what our code uses to find out what the item type is. Then, when we use the call(void *, ...) function, it counts how many variables there are - this is how many extra arguments the call function should have been passed - then replaces the variables with the values we've called it with, then does the calculations.
The other option (which will probably be considerably faster and easier to extend) is to use something like libjit to do most of that work for you. I've never used it, but a JIT compiler gives you some basic building blocks (like add, multiply, etc. "instructions") that you can string together as you need them, and it compiles them down to actual assembly code (so no going through a constructed syntax tree calling function pointers like we had to before) so that when you call it it's as fast and dynamic as possible.
I don't know libjit's API, but it looks easily capable of doing what you seem to need. The make_func and free_func could all be pretty much the same as they are above (you might have to alter your calls to call_func) and would basically construct, use, and destroy a JIT object based on how it parses the user's string. The same as above, really, but you wouldn't need to define the syntax tree, data types, etc. yourself.
Hope that is somewhat helpful.
libtcc (from TCC) can be used as a very small and fast JIT; see libtcc_test.cc for sample usage.
Related
I've made my own implementation of a HashMap/HashTable (i know they are different but it's irrelevant for this question).
In this implementation, I'd like it to be very flexible. I want to be able to store ints, structs, chars, strings, etc all as keys or values without having to change the code of my algorithms. For example, in Java I can just do:
HashMap<Integer, MyPersonalClass> and it will just work. In C, I know there is no direct equivalent except void*. The issue is, if I have:
/* Node structure. */
struct hm_Node
{
void *key, *value;
struct hm_Node *next;
};
As the Node(s) that make up my HashMap/HashTable, then my hash() method needs to somehow parse the key correctly. So far I've only looked up an algorithm for char*.
Is there something like:
// This may not be valid code, just using it as an example
unsigned int hash(void *ptr)
{
switch(typeof(ptr)) // I know ptr is of type void*
{
case char*: ... break;
case char: ... break;
case int: ... break;
}
}
How does that work exactly? I'm just trying to avoid having a whole different implementation for a HashMap of X, Y, and Z types. Thanks.
Look at the implementation of, for example qsort:They let the user provide the comparison function in order to be able to implement arbitrary sorts.
You can go the same way by letting the user provide a proper hash function through a function pointer - If you want, you can supply them with some pre-built hash functions for standard types they can re-use.
Mostly for fun, I've decided to write my own minimal test framework for my C code. I use a basic struct for the test information, create an array of test structs and then iterate over them to run all the tests. This amounts to a very small amount of work for a fairly elegant (imho) solution.
However, the one thing that is a little annoying is that I cannot figure out how to define functions as function pointers instead of defining the function and then creating a function pointer later.
I have the following (which works just fine):
typedef int (* test_p) (void);
struct test {
char * desc;
test_p func;
};
int
example_test (void) {
puts("This is a test");
return 0;
}
void
run_test (char * test_name, test_p test) {
printf("Testing %s\t\t\t[ PEND ]\r", test_name);
char * test_result = (test() ? "FAIL" : "PASS");
printf("Testing %s\t\t\t[ %s ]\n", test_name, test_result);
}
int
main (void) {
struct test test_list [] = {
{ "example test", (test_p )example_test }
};
for ( int i = 0; i < 1; i ++ ) {
run_test(test_list[i].desc, test_list[i].func);
}
return 0;
}
However, I am hoping I can remove the need for the casting in the struct and instead define the function as being a function pointer from the beginning. The following is an example of how I would like this to work (assuming many of the same things as above):
test_p
example_test = {
puts("This is a test");
return 0;
}
If I could do something like this, then in the struct, I could simply have the func field be example_test rather than (test_p )example_test. Is this (or something like it) possible? If not, is there a reason why not (If that reason is simply "because it wasn't added to the language", that's fine)?
A function pointer is one kind of thing and a function is another kind of thing so you can't really make the latter be the former. But if you use a function name where a function pointer is expected, that produces a pointer to the function, so you can just remove the unnecessary cast, as WhozCraig said in the first comment above. You write
If I could do something like this, then in the struct, I could simply have the func field be example_test rather than (test_p )example_test.
You can do that, with example_test defined just as it is in your current code ... did you try that?
You can also forward declare a function, like so:
typedef int test_func(void); // note no indirection
typedef test_func* test_p;
test_func example_test;
It would be nice if you could use that sort of syntax when you define the function, as in your attempted syntax, but there's simply no way to do that in C ... you have to explicitly provide the return type and parameter list.
Another detail is that, when you invoke the function pointed to by a function pointer, you don't have to dereference it ... that's why you were able to write
test()
instead of
(*test)()
although the latter also works. (In fact, because the deference is stripped, (********test)() also works ... but only do that if you're trying to win an obfuscation contest.)
What you are describing is a kind of meta-programming. Rather than writing code to explicitly solve the problem, you are concerned with a kind of syntactic structure that will allow you to define a whole raft of test functions without unnecessary cruft.
In Lisp you would use macros. In C++ you might use templates and/or lambdas. In C you use macros.
So you need to write a macro that:
takes a name and descriptive text as arguments
defines a static variable of type function (created from that name using token pasting)
defines a function (using a name created by token pasting)
[edit] At this point you have achieved the goal: you have created the function and given it a name that is (only) a function pointer, and you can use that name in your struct without a cast. I would suggest one additional step, the macro also:
adds the variable/function and descriptive text to a list of functions to be tested.
Then your boilerplate loop iterates over the structure calling each function and reporting the results using the descriptive text. Problem solved.
Some people don't like macros, but they are ideally suited to this situation, and there is no other way to do it in C. I did something just like this before making the move to C++.
Sometimes, in C, you do this:
typedef struct foo {
unsigned int some_data;
} foo; /* btw, foo_t is discouraged */
To use this new type in an OO-sort-of-way, you might have alloc/free pairs like these:
foo *foo_alloc(/* various "constructor" params */);
void foo_free(foo *bar);
Or, alternatively init/clear pairs (perhaps returning error-codes):
int foo_init(foo *bar, /* and various "constructor" params */);
int foo_clear(foo *bar);
I have seen the following idiom used, in particular in the MPFR library:
struct foo {
unsigned int some_data;
};
typedef struct foo foo[1]; /* <- notice, 1-element array */
typedef struct foo *foo_ptr; /* let's create a ptr-type */
The alloc/free and init/clear pairs now read:
foo_ptr foo_alloc(/* various "constructor" params */);
void foo_free(foo_ptr bar);
int foo_init(foo_ptr bar, /* and various "constructor" params */);
int foo_clear(foo_ptr bar);
Now you can use it all like this (for instance, the init/clear pairs):
int main()
{
foo bar; /* constructed but NOT initialized yet */
foo_init(bar); /* initialize bar object, alloc stuff on heap, etc. */
/* use bar */
foo_clear(bar); /* clear bar object, free stuff on heap, etc. */
}
Remarks: The init/clear pair seems to allow for a more generic way of initializing and clearing out objects. Compared to the alloc/free pair, the init/clear pair requires that a "shallow" object has already been constructed. The "deep" construction is done using init.
Question: Are there any non-obvious pitfalls of the 1-element array "type-idiom"?
This is very clever (but see below).
It encourages the misleading idea that C function arguments can be passed by reference.
If I see this in a C program:
foo bar;
foo_init(bar);
I know that the call to foo_init does not modify the value of bar. I also know that the code passes the value of bar to a function when it hasn't initialized it, which is very probably undefined behavior.
Unless I happen to know that foo is a typedef for an array type. Then I suddenly realize that foo_init(bar) is not passing the value of bar, but the address of its first element. And now every time I see something that refers to type foo, or to an object of type foo, I have to think about how foo was defined as a typedef for a single-element array before I can understand the code.
It is an attempt to make C look like something it's not, not unlike things like:
#define BEGIN {
#define END }
and so forth. And it doesn't result in code that's easier to understand because it uses features that C doesn't support directly. It results in code that's harder to understand (especially to readers who know C well), because you have to understand both the customized declarations and the underlying C semantics that make the whole thing work.
If you want to pass pointers around, just pass pointers around, and do it explicitly. See, for example, the use of FILE* in the various standard functions defined in <stdio.h>. There is no attempt to hide pointers behind macros or typedefs, and C programmers have been using that interface for decades.
If you want to write code that looks like it's passing arguments by reference, define some function-like macros, and give them all-caps names so knowledgeable readers will know that something odd is going on.
I said above that this is "clever". I'm reminded of something I did when I was first learning the C language:
#define EVER ;;
which let me write an infinite loop as:
for (EVER) {
/* ... */
}
At the time, I thought it was clever.
I still think it's clever. I just no longer think that's a good thing.
The only advantage to this method is nicer looking code and easier typing. It allows the user to create the struct on the stack without dynamic allocation like so:
foo bar;
However, the structure can still be passed to functions that require a pointer type, without requiring the user to convert to a pointer with &bar every time.
foo_init(bar);
Without the 1 element array, it would require either an alloc function as you mentioned, or constant & usage.
foo_init(&bar);
The only pitfall I can think of is the normal concerns associated with direct stack allocation. If this in a library used by other code, updates to the struct may break client code in the future, which would not happen when using an alloc free pair.
I want to do some object-oriented style programming in C using polymorphism, where my interface class contains a pointer to a table of functions. Example something like:
/* Implement polymorphism in C, Linux kernel-style */
struct statement {
const struct statement_ops *ops;
struct list_head list; /* when on master input list */
void *private; /* pointer to type-specific data */
};
struct statement_ops {
int (*analyse)(void *private, int pc);
int (*get_binary_size)(void *private);
};
void user(void)
{
struct statement *s = make_a_statement();
if (s->ops->analyse(s->private, foo))
blah blah;
}
I'd like to be able to write something without explicitly passing s->private into every "method". Any ideas? Some macro tricks maybe?
If this is part of the public interface, you can add accessor functions. A hidden benefit is that you can do sanity checks and other work in the accessor. (Note I called the "this" pointer "o", as in "object". I prefer it that way for consistency.)
int statement_analyse (struct statement *o, int pc)
{
assert(pc >= 0);
int ret = o->ops->analyse(o->private, pc);
assert(ret >= 0);
return ret;
}
You can now call this without the explicit passing of "private".
void user(void)
{
struct statement *s = make_a_statement();
if (statement_analyse(s, foo))
blah blah;
}
While it may seem that this provides no benefit, because you still have to implement the accessors, assuming that you want a well defined and robust interface, the accessor functions are the only sane place to put the assertions and the interface documentation. In fact, if you write good assertions, the assertions themselves help document the interface. And once you add sanity checks in the accessors, you don't have to add them in the actual methods they call.
Of course, this approach only makes sense when the function called via the function pointer will be something provided by the user, or in some other way can be different things. If there's a single analyse() method that will always do the same thing, you can simply implement a statement_analyse() that directly does what it needs to do.
Small note: when doing OOP, I prefer to typedef the structs and give them CamelCase names. I use this convention as a way of telling that the struct is opaque and should only be accessed via its public interface. It also looks nicer, though that is subjective. I also prefer having the user allocate the memory for the struct itself, as opposed to the constructor malloc'ing it. That avoids having to handle malloc failure, and makes the program a little bit more efficient.
typedef struct {
...
} Statement;
void Statement_Init (Statement *o);
int Statement_Analyse (Statement *o, int pc);
Unfortunately, writing your methods to allow the passing of a self or this object is the only way to achieve this in C.
You can use macro tricks to hide part of it, but at that point it's not really C any more.
As the other answers say, there is no way to do this without calling the function with the appropriate pointer, but (as Williham Totland suggests) you could use macros to streamline the calls (requires a compiler with variadic macro support):
// macro_call.c
#define C_ARGS(stmnt, func, ...) (stmnt)->ops->func((stmnt)->private, ...)
#define C_NOARGS(stmnt, func) (stmnt)->ops->func((stmnt)->private)
C_ARGS(s, analyse, 1);
C_ARGS(s, lots_of_args, 1, 2, 3, 4);
C_NOARGS(s, no_args);
(The C is for "call".)
Doing the preprocessing on that (via gcc -E macro_call.c) gives:
(s)->ops->analyse((s)->private, 1);
(s)->ops->lots_of_args((s)->private, 1, 2, 3, 4);
(s)->ops->no_args((s)->private);
This is similar to the accessor function version: the macro version is slightly more flexible in some ways, but it is also less safe and could lead to subtle errors and mistakes.
There are two macros because passing no extra arguments to C_ARGS would result in s->ops->func(s->private, ), I think it is possible to fix this, but it is awkward and would require significantly more code (empty __VA_ARGS__ are notoriously hard to deal with).
Is there a "proper" way to implement higher order functions in C.
I'm mostly curious about things like portability and syntax correctness here and if there are more than one ways what the merits and flaws are.
Edit:
The reason I want to know how to create higher order functions are that I have written a system to convert PyObject lists (which you get when calling python scripts) into a list of C structures containing the same data but organized in a way not dependant on the python.h libraries. So my plan is to have a function which iterates through a pythonic list and calls a function on each item in the list and places the result in a list which it then returns.
So this is basically my plan:
typedef gpointer (converter_func_type)(PyObject *)
gpointer converter_function(PyObject *obj)
{
// do som stuff and return a struct cast into a gpointer (which is a void *)
}
GList *pylist_to_clist(PyObject *obj, converter_func_type f)
{
GList *some_glist;
for each item in obj
{
some_glist = g_list_append(some_glist, f(item));
}
return some_glist;
}
void some_function_that_executes_a_python_script(void)
{
PyObject *result = python stuff that returns a list;
GList *clist = pylist_to_clist(result, converter_function);
}
And to clearify the question: I want to know how to do this in safer and more correct C. I would really like to keep the higher order function style but if that is frowned upon I greatly appreciate ways to do this some other way.
Technically, higher-order functions are just functions that take or return functions. So things like qsort are already higher-order.
If you mean something more like the lambda functions found in functional languages (which is where higher order functions really become useful), those are quite a bit harder and can't be done naturally in current standard C. They're just not part of the language. Apple's blocks extension is the best candidate. It only works in GCC (and LLVM's C compiler), but they are really useful. Hopefully something like that will catch on. Here's a few relevant resources:
Apple's documentation on the feature (references some Apple-specific technologies and also addresses Objective-C, but the core block stuff is part of their extensionto C)
Here's a good intro on blocks
Cocoa for Scientists' overview of C blocks
The big problem with implementing higher-order functions in C is that to do anything non-trivial you need closures, which are function pointers augmented with data structures containing local variables they have access to. Since the whole idea behind closures is to capture local variables and pass those along with the function pointer, it's hard to do without compiler support. And even with compiler support it's hard to do without garbage collection because variables can exist outside of their scope, making it hard to figure out when to free them.
This is an answer to the question: how to compose functions in C, which is redirected here.
You can create a data structure to implement a list data type.
that structure can contain function pointers.
#include<stdlib.h>
#include<malloc.h>
typedef (*fun)();
typedef struct funList { fun car; struct funList *cdr;} *funList;
const funList nil = NULL;
int null(funList fs){ return nil==fs; }
fun car(funList fs)
{
if(!null(fs)) return fs->car;
else
{
fprintf(stderr,"error:can't car(nil) line:%d\n",__LINE__);
exit(1);
}
}
funList cdr(funList ls)
{ if(!null(ls)) return ls->cdr;
else
{
fprintf(stderr,"error:can't cdr(nil) line:%d\n",__LINE__);
exit(1);
}
}
funList cons(fun f, funList fs)
{ funList ls;
ls=(funList) malloc(sizeof(struct funList));
if(NULL==ls)
{
fprintf(stderr,"error:can't alloc mem for cons(...) line:%d\n",__LINE__);
exit(1);
}
ls->car=f;
ls->cdr=fs;
return ls;
}
we can write a function comp which applies a list of functions:
type_2 comp(funList fs, type_1 x)
{
return (null(fs)) ? x : car(fs)(comp(cdr(fs),x));
}
An example of how it works. We use (f g h) as a short notation for cons(f,cons(g,cons(h,nil))), which is applied to a given argument x:
comp((f g h),x)
=
f(comp((g h),x))
=
f(g(comp((h),x)))
=
f(g(h(comp(nil,x))))
=
f(g(h(x)))
if you had used the polymorphic list type in a typed language like SML or Haskell the type of comp should be:
comp :: ([a -> a],a) -> a
because in that context all the members in a list have the same type.
C can be more flexible in this sense. Maybe something like
typedef void (*fun)();
or
typedef (*fun)();
you should see what the C manual say about this. And be sure that all contiguous functions have compatible types.
The functions to compose should be pure, i.e. without side effects nor free variables.
In straight c, this is really only done through function pointers, which are both a pain and not meant for this type of thing (which is partially why they are a pain). Blocks (or closures, according to non-apple) are fantastic for this, though. They compile in gcc-4.x or something, and icc something, but regardless thats what you're looking for. Unfortunately, I can't seem to find any good tutorials online, but suffice to say it works something like this:
void iterate(char *str, int count, (^block)(str *)){
for(int i = 0; i < count; i++){
block(list[i]);
}
}
main() {
char str[20];
iterate(str, 20, ^(char c){
printf("%c ", c);
});
int accum = 0;
iterate(someList, 20, ^(char c){
accum += c;
iterate(str, 20, ^(char c){
printf("%c ", c);
});
});
}
obviously this code is pointless, but it it prints each character of a string (str) with a space in between it, then adds all of the characters together into accum, and every time it does it prints out the list of characters again.
Hope this helps. By the way, blocks are very visible in Mac OS X Snow Leopard api-s, and I believe are in the forthcoming C++0x standard, so they're not really that unusual.
If you're keen on doing this in plain C, you need to remember to include the option to pass in a context pointer from the caller of the functor (the higher-order function) to the function passed in. This lets you simulate enough of a closure that you can make things work easily enough. What that pointer points to... well, that's up to you, but it should be a void* in the functor's API (or one of the many aliases for it, such as gpointer in the GLib world or ClientData in the Tcl C API).
[EDIT]: To use/adapt your example:
typedef gpointer (converter_func_type)(gpointer,PyObject *)
gpointer converter_function(gpointer context_ptr,PyObject *obj)
{
int *number_of_calls_ptr = context_ptr;
*number_of_calls_ptr++;
// do som stuff and return a struct cast into a gpointer (which is a void *)
}
GList *pylist_to_clist(PyObject *obj, converter_func_type f, gpointer context_ptr)
{
GList *some_glist;
for each item in obj
{
some_glist = g_list_append(some_glist, f(context_ptr,item));
}
return some_glist;
}
void some_function_that_executes_a_python_script(void)
{
int number_of_calls = 0;
PyObject *result = python stuff that returns a list;
GList *clist = pylist_to_clist(result, converter_function, &number_of_calls);
// Now number_of_calls has how often converter_function was called...
}
This is a trivial example of how to do it, but it should show you the way.
Practically any interesting higher order function application requires closures, which in C entails the laborous and error-prone routine of manually defining and filling struct function arguments.
It's very difficult to do in straight C. It's more possible in C++ (see functors tutorial or Boost's bind and function libraries). Finally, C++0x adds native support for lambda functions, which takes care for you of capturing in closure all of the variables that your funcion depends on.
If you want to create higher order functions, don't use C. There are C solutions to your problem. They may not be elegant, or they may be more elegant that you realize.
[Edit] I suggested that the only way to achieve this was to use a scripting language. Others have called me out on it. So, I'm replacing that suggestion with this: [/Edit]
What are you trying to achieve? If you want to mimic closures, use a language that supports them (you can tie into Ruby, lua, javascript, etc through libraries). If you want to use callbacks, function pointers are ok. Function pointers combine the most dangerous areas of C (pointers and the weak type system), so be careful. Function pointer declarations are not fun to read, either.
You find some C libraries using function pointers because they have to. If you're writing a library, maybe you need to use them, too. If you're just using them within your own code, you're probably not thinking in C. You're thinking in lisp or scheme or ruby or ... and trying to write it in C. Learn the C way.